This week, I decided to do a Q&A of AI related questions I have received on substack, Twitter, discord and forum.

Open source models running on domestic AI chip. New way of measuring from Stepfun.

I talked about AI robots, AI applications in industries and running AI on different type of chips.

I will have more on China and AI once I get back from my China trip. The way that AI is incorporated into Chinese society since DeepSeek is something to behold.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tphuang.substack.com

This week, I invited my friend Simon on the show to talk about AI infrastructure. This is a very interesting topic to think about. We talked about the reason for Nvidia’s market dominance and the problems that Intel and AMD have faced in developing their own market place. We talked about the number of chips involved and the challenges in building large data centers. Here is a chart of the Hopper deliveries for 2024. Keep in mind that Google also has their own TPU for inference.

Here is a chart of Chinese Hyperscaler’s purchases of H20 in 2024. But keep in mind that they do rent compute from public/state built data centers that use smuggled in H100/H200 as well as Ascend chips.

Here is the deliveries of H100/H800 in 2023. So if we just consider the Hopper deliveries from 2023 and 2024, the American hyperscalers have quite the computation advantage over Chinese hyperscalers. Although, the gap is much smaller if we factor in all the smuggled in chips. As I’ve discussed before, China had them everywhere to the point where they were sitting idle in many cases.

I also asked Simon about the amount of chips needed to do inference. How was Tencent able to have enough compute for 8 million DeepSeek R1 requests at the same time with much less compute than what Google had. He sent me this chart of where the results improve logarithmically in compute for reasoning. So if you look at green line, going from 4 generations to 16 generations improved accuracy by about 15% and then going from 16 generations to 64 generations improved the result by another 5%. If we look at Google search results, they appear to be giving progressively better AI results on top. That is likely from running through and generation more tokens on their reasoning models.

So you can choose to serve 8 million prompts at same time with vastly less compute, but you will also have inferior results. although you will also reach diminishing returns pretty soon.

Here is a chart of rental cost from various Nvidia chips in China. Despite the increased demand for inference post DeepSeek, the cost of all Nvidia rental cost continue to drop, so they are likely not facing a crunch for computation yet.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tphuang.substack.com

I was inspired to make an episode on manufacturing and why it is difficult to do manufacturing in America after I watched this video, I asked Lei to join me to discuss how China has taken over the expertise on manufacturing all the “basic stuff”

Of course, China has now developed the ability to manufacture higher end stuff and increasingly designing its own products. But China’s upstream supply chain dominance is something that’s not well understood.

So in this episode, we went over our experience with manufacturing and supply chain. Lei is more familiar with this than me, so he did most of the talking.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tphuang.substack.com

For the third episode, I invite Glenn Luk onto the show to talk China’s vast HSR. He helped us debunk many of the myths regarding China’s HSR economics. We also talked about how HSR helps facilitate travel, improve Quality of Life and increase freight inside China.

Here is a look at the current Chinese high speed rail map

And its overall railway map. Very well connected and you can imagine that many lines end up in more remote areas that cannot justify the expense of building HSR. Even so, they are still likely faster than the 70km/h train that I used to take growing up.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tphuang.substack.com

This week, I welcomed Snek onto my podcast to talk about his thoughts on the ongoing Russian/Ukraine conflict. This is an area that I have not followed closely, but Snek is very well versed in. We recorded with lower volume, so raise the volume to hear it clearly.

We look at the conflict in 3 areas:

* The current state of the conflict.

* The usage of drones and AI in this conflict

* The industrial aspect of this conflict and why US/Europe’s military industrial complex have struggled keeping up.

Although this veers off my usual focus on Chinese tech, I think this conflict does have significant implications on 2 of the areas that I really enjoy looking into: drones and industrial capacity.

As a follow up, Snek pointed me to this article:

Pentagon only has 1/4 of the Patriot missiles it needs. Ukraine and the Middle East have consumed all of this. America is seriously running low on this stuff and it will likely burn up even more over the continued conflicts.

On the topic of longer ranged attack drones and their motor. You can see just how cheap China has gotten down the cost of micro turbojet. I’d imagine all the low cost long range drones or missiles around the world are made with these engines.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tphuang.substack.com

I invited Lei on the show for our first episode

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tphuang.substack.com